Table 4 Tunable hyperparameters in the... | Emerald Publishing

Table 4

Tunable hyperparameters in the CNN architecture and training pipeline

Parameter	Description	Default value
block_ch	Number of filters in block pooling layer	256
block_size	Size/stride of block pooling kernel	100
conv_chs	List of channels in conv layers	[256, 128, 64, 32, 16, 8]
fc_dims	Sizes of FC hidden layers	[512, 256, 64]
dropout_rate	Dropout probability in FC layers	0.4
learning_rate	Learning rate for Adam optimizer	1 × 10^–3
batch_size	Number of samples per batch	128
num_epochs	Total number of training epochs	120
val_split	Validation set fraction	0.15
test_split	Test set fraction	0.15
output_dim	Dimension of output (R_e, R_ν)	2
in_channels	Input channels (geometry and features)	2

Parameter	Description	Default value
block_ch	Number of filters in block pooling layer	256
block_size	Size/stride of block pooling kernel	100
conv_chs	List of channels in conv layers	[256, 128, 64, 32, 16, 8]
fc_dims	Sizes of FC hidden layers	[512, 256, 64]
dropout_rate	Dropout probability in FC layers	0.4
learning_rate	Learning rate for Adam optimizer	1 × 10^–3
batch_size	Number of samples per batch	128
num_epochs	Total number of training epochs	120
val_split	Validation set fraction	0.15
test_split	Test set fraction	0.15
output_dim	Dimension of output (R_e, R_ν)	2
in_channels	Input channels (geometry and features)	2